About

I’m looking for an experienced Stable Diffusion LoRA trainer to create a high-quality black-and-white comic lineart LoRA using a dataset of around 160 clean PNG images. The style is defined by heavy inks, dramatic shadow shapes, bold line weight variation, intense cross-hatching, and highly dynamic superhero-style anatomy. The goal of this LoRA is to accurately capture the inking style and rendering language so it can generate new characters, splash pages, dramatic portraits, and dynamic figure illustrations in this same bold black-and-white comic style. I am not looking to recreate full panel layouts or page structures. The focus is purely on style transfer, line quality, dramatic shading, and expressive figure drawing. Some images in my dataset are very large (up to 10k–20k pixels), so I need someone familiar with handling high-resolution artwork and resizing it properly for LoRA training without destroying line integrity or detail. Aspect-ratio-preserving bucket resizing is preferred; strict cropping is not required since composition learning is not the main priority — capturing the style is. I need a trainer who is proficient with kohya-ss / sd-scripts, understands LoRA hyperparameters, and has experience training models specifically for lineart, manga, or comic-style rendering. You should know how to choose appropriate dim/alpha settings, learning rates, repeats, and augmentation choices to maintain detail without overfitting. I want the resulting LoRA to generalize well and consistently produce bold, expressive black-and-white artwork. Deliverables include the final .safetensors LoRA file, a brief README describing your training settings and recommended inference parameters (weights, samplers, resolutions), and 5–10 sample outputs showing characters or splash-page style images generated using the finished LoRA. A ComfyUI testing workflow is a plus. My budget is $100–$250, depending on experience and output quality. When submitting a proposal, please include samples of LoRAs you’ve trained before (especially lineart/comic styles), your expected workflow, timeline, pricing, and a short explanation of how you avoid overfitting during LoRA training.